The current experiments on structural determination cannot keep up the pace with the steadily emerging RNA sequences and new functions. This underscores the urgent request for an accurate free energy model for RNA tertiary folds, from which one can predict structures from sequences. Furthermore, there is increasing support for the idea that large structured RNAs may adopt a variety of conformational states, rather than just one, during the course of performing its biological function, this is particularly so in the replication of RNA viruses and a central tenet of riboswitch-mediated regulation of gene expression in bacteria. Although considerable progress has been made in mechanistic studies, accurate prediction for RNA tertiary folding from sequence remains an unsolved problem. The first and most important requirement for understanding and predicting of RNA folding from RNA structural fluctuations to large conformational changes is an accurate free energy model. Support from this grant has allowed us to develop a novel virtual bond-based RNA free energy model that enables much better predictions than other existing models for simple tertiary structures (pseudoknots). We now propose to go beyond the simple pseudoknots by studying all-atom, larger, more complex RNA tertiary folds. Our approach will be based on rigorous, first principles analytical calculations. A key advantage of the approach is the completeness and certainty in conformational sampling (entropy). Incorrect entropy results in poor predictions. Preliminary tests using experimental data have shown significant improvements from our approach in both accuracy and specificity than existing folding algorithms. The success attests the high promise of the new approach proposed in this grant. Our specific aims are: (a) Systematic model development for tertiary folding free energies. (b) Developing a novel approach for complex, larger tertiary folds. (c) Developing a 3D all-atom model. (d) Systematic test and refinement of the model using experimental structural data.